US 7783425 B1 Resumen An integrity monitoring method for an aircraft is disclosed. The integrity monitoring method includes determining a set of non-uniform weights. The non-uniform weights are based on a least squares approximation of a linearized measurement equation. The integrity monitoring method also includes applying the non-uniform weights in a receiver autonomous integrity monitoring (RAIM) system. Further, the integrity monitoring method includes determining a reduced integrity limit based on the output of the RAIM system.
Reclamaciones(16) 1. An integrity monitoring system for an aircraft, comprising:
a weight generation subsystem, generating a set of non-uniform weights based on a least squares approximation of a linearized measurement equation;
a receiver autonomous integrity monitoring (RAIM) subsystem applying the set of non-uniform weights; and
a horizontal integrity limit (HIL) subsystem generating a reduced horizontal integrity limit based on the output of the RAIM subsystem.
2. The system of
3. The system of
4. The system of
5. The system of
a HIL evaluation subsystem.
6. The system of
7. An integrity monitoring system, comprising:
matrix including a set of non-uniform weights, based on a least squares approximation of a linearized measurement equation;
a receiver autonomous integrity monitoring (RAIM) system for applying the set of the non-uniform weights; and
an integrity limit subsystem for determining a reduced integrity limit based on the output of the RAIM system.
8. The system of
9. The system of
10. The method of
11. The method of
a means for applying a look-up table method for evaluation of the integrity limit.
12. An autonomous integrity monitoring system for use in a receiver, comprising:
a weight generation subsystem for determining a set of non-uniform weights, based on a least squares approximation of a linearized measurement equation;
a weight application subsystem for applying the non-uniform weights; and
an integrity limit subsystem for providing a reduced integrity limit based on an output of the weight application subsystem.
13. The system of
14. The system of
15. The system of
16. The system of
a look-up subsystem for applying a look-up table method for evaluation of the integrity limit.
Descripción This application is related to U.S. patent application Ser. No. 11,170,665 entitled Integrity-Optimized Solution Separation Method for Inertially-Aided Aircraft Autonomous Integrity Monitoring, filed on the same day herewith incorporated by reference in its entirety. A major shortcoming of Receiver Autonomous Integrity Monitoring (RAIM) Fault Detection and Exclusion (FDE) is widely recognized to be it occasional lack of availability, in the sense that the integrity level it can protect in any given situation cannot always be smaller than the required alert limit for a given flight operation. Ever since the RAIM FDE algorithms were developed, accepted and published more than ten years ago, the system availability has been bound, by and large, to the limitations of its monitoring mechanism to be able to detect and exclude a single measurement fault. There have been many studies made on evaluating the system availability, usually on a global basis, and they have considered various assumptions of constellation size, horizon mask angle, and aiding with baro-altitude, inertial, and additional non-GPS satellites, or even with Loran. In all of these studies, however, the basic shortcomings of RAIM FDE remained essentially bound by its theoretical limits. More recently, there has been a proposition that departs from conventional assumptions made in the standard RAIM-FDE by better balancing accuracy and integrity levels. This proposition results in the lowering of the horizontal integrity limit for all satellite measurement geometries, thereby improving system availability in the process. Accordingly, there is a need for a methodology called Novel Integrity Optimized RAIM ( It would be desirable to provide a system and/or method that provides one or more of these or other advantageous features. Other features and advantages will be made apparent from the present specification. The teachings disclosed extend to those embodiments which fall within the scope of the appended claims, regardless of whether they accomplish one or more of the aforementioned needs. What is provided is an integrity monitoring method. The method comprises determining a set of non-uniform weights based on a least squares approximation of a linearized measurement equation. The method also comprises applying the non-uniform weights in a receiver autonomous integrity monitoring (RAIM) system. Further, the method comprises determining a reduced integrity limit based on the output of the RAIM system. What is also provided is an integrity monitoring system for an aircraft. The integrity monitoring system comprises a weight generation subsystem. The weight generation subsystem generates a set of non-uniform weights based on a least squares approximation of a linearized measurement equation. The system also comprises a receiver autonomous integrity monitoring subsystem applying the set of non-uniform weights. Further, the system comprises a horizontal integrity limit subsystem generating a reduced horizontal integrity limit based on the output of the RAIM subsystem. Further, what is provided is an integrity monitoring system. The integrity monitoring system comprises a means for determining a set of non-uniform weights based on a least squares approximation of a linearized measurement equation. The integrity monitoring system also comprises the means for applying the non-uniform weights in a receiver autonomous integrity monitoring system. Further, the integrity monitoring system comprises a means for determining a reduced integrity limit based on the output of the RAIM system. Alternative examples and other exemplary embodiments may also be provided which relate to other features and combination of features as may be generally recited in the claims. The invention will become more fully understood from the following detailed description, taken in conjunction with the accompanying drawings, wherein like reference numerals refer to like elements, in which: Before describing in detail the particular improved system and method, it should be observed that the invention includes, but is not limited to a novel structural combination of conventional data/signal processing components and communications circuits, and not in the particular detailed configurations thereof. Accordingly, the structure, methods, functions, control and arrangement of conventional components and circuits have, for the most part, been illustrated in the drawings by readily understandable block representations and schematic diagrams, in order not to obscure the disclosure with structural details which will be readily apparent to those skilled in the art, having the benefit of the description herein. Further, the invention is not limited to the particular embodiments depicted in the exemplary diagrams, but should be construed in accordance with the language in the claims. To demonstrate the general approach of In general, the An analysis of weighted least-squares with the basic linearized measurement equation where there may be an overdetermined system of linear equations is begun with:
where z is the measurement, H providing the idealized linear connection, and x is the unknown state vector, and ε is the measurement error vector. Assume, as is usual in the RAIM problem after normalization for possible unequal measurement variances, that the random elements of ε, as shown in Eq. (1) are uncorrelated and have equal variances. The weighted least-squares estimate is then:
Assume V to be diagonal in this analysis, and the square root of V describes the weights given to the measurements, i.e.,
where w is the square matrix containing positive weights along the diagonal. Of course, the estimate given by Eq. (2) reduces to the usual optimal least-squares estimate when all the elements of w (or V) are equal. Otherwise, the estimate is suboptimal. Of primary interest here is the suboptimal estimate. The test statistic will be the usual parity vector, and it is formed by
where P is the parity transformation matrix associated with H. In summary, assume that the estimate and test statistic are given by the combined matrix equation:
It is easily verified by noting that P is orthogonal with H orthogonal with H that the estimation the estimation error of x and parity vector p are related to ε through the equations:
It is important to note that e and p are functions of ε alone (and not x), and this is as it must be in the RAIM problem. Also, considering just the random part of e and p, it is easily verified that the covariances are given by the equations:
where n is the number of measurements and σ The presence of w in the x-error equation, Eq. (7), complicates the weighted RAIM analysis considerably. Forming the cross-covariance between p and e, in general, all the components of p and e are mutually correlated. This causes the scatter plot of the horizontal position error versus the norm of p (i.e. the detection test statistic) to be skewed as shown in In view of the previous discussion, and noting that the parity vector and associated parity space are invariant with the weights, one might logically ask, “From whence comes the RAIM improvement?” The answer is this: by judicious choice of w, the deterministic position error caused by the bias on the most-difficult-to-detect satellite can be reduced significantly. This, in turn, reduces the slope line for that particular satellite, and thus the horizontal integrity limit (HIL) is lowered accordingly. This is done at some expense in terms of rms error due to noise, but this may be acceptable for improved availability. It is also worth mentioning that the foregoing remarks apply only to the detection half of RAIM. If the parity method is used for identification, then the identification half of our proposed The true benefit of the In RAIM, the worst case satellite is generally associated with the largest slope. To account for the noise scatter, additional adjustments were needed in the usual RAIM, but this was relatively uncomplicated because the noise scatter distribution is uncorrelated between its two dimensions of parity and error. The same is not true, in general, for The cross-covariance between the parity and the position error vectors, from Eqs. (6) and (8), may be formed as
For the original RAIM, where V=I, Cov(p, e)=0 because P is orthogonal with H. For (a) Monte Carlo: This is simply a “brute-force” exercise in running a massive number of randomly-generated measurements while simulating a varying-size fault and counting the samples that fall within the missed detection quadrant to establish the Integrity Level Y (see Note, from While this Monte Carlo method may be too computationally taxing for real-time processing, it is straightforward and defensible, and useful for offline validation purposes. (b) Upper Bound: A more analytical alternative to determining the integrity level may be to compute a conservative upper bound for it. The missed detection probability is to be contained not just within the upper left quadrant, but within both the upper left and upper right quadrants (shaded region of distribution in It may also be desirable to consider the nontrivial correlation that exists between the parity and error variates. A correlation to a first-order approximation may be established. First, the angle of the ellipse in This approach, while quite amenable to real-time processing, produces somewhat pessimistic results. Therefore, another alternative that proves to be quite practical, once validated may be applied. (c) Lookup Table: The notion of a lookup table conjures the idea of some compromise between rigorous computation that has to be computed and validated offline, so that this table can be referenced online with minimal processing. Every lookup table requires a definition of the set of parameters that would fully describe the relevant operating conditions. To simplify this aspect of the lookup table, it may be assumed that the distribution is, to a first-order approximation, bivariate Gaussian (see 1. Number of satellites (affects the Threshold); 2. Variance of the Error dimension c 3. Correlation coefficient of the bivariate Gaussian distribution (related to c 4. Slope of the mean of the bivariate Gaussian distribution. The computation of the covariance matrix C In offline processing, the integrity limit may be computed for a set of cardinal points of a 4-dimensional array that represents the lookup table. In online processing, an interpolation process among the appropriate neighboring cardinal points may be carried out to compute the integrity limit. Note that the mean of the bivariate Gaussian distribution does not have to be dealt with as two separate parameters (μ Methods for seeking a weighting matrix that would optimize the integrity limit is described. (a) Monte Carlo Approach In a Monte Carlo approach, the different values of the diagonal weighting matrix w are randomized, the resultant integrity limit is evaluated, and this is repeated for a large number of trials, ultimately choosing the set of weights that yield the lowest integrity limit. The success of finding a good set of weights yielding the lowest integrity limit depends considerably on the sample size since its performance is ultimately statistical in nature. This approach is quite computationally intensive because of the dimensionality of w. (b) Iterative Search A more efficient search may well be based on any one of numerous methods available from the catalog of optimization science, but they all face the difficulties and challenges of what amounts to solving a global optimization problem with multiple local extrema. What may be more valuable is an ad hoc iterative search method that is straightforward and appears to do a reasonably good job of producing a solution that is close to the optimum. An initial search point may be chosen as unity weights. With each iteration, the following is computed: (a) the variance of the random position error, (b) correlation coefficient of the bivariate distribution in the parity-position scatter plot, and (c) slopes for each of the satellites in the measurement suite (see the Appendix for details). Using Items (a), (b) and (c) as input parameters to the Lookup Table, the horizontal integrity level L Heuristically, a larger protection level L Extension of Search In each search, it may be desirable to terminate the iterative process when the integrity limit (the largest of the L The ultimate utility of In a typical analysis conducted with tools developed for standard RAIM-FDE at Rockwell-Collins, Inc. of Cedar Rapids, Iowa, to certify system availability performance of its commercial avionics GPS receiver products, the following availability rates were obtained (this assumes an absence of selective availability, and visibility of satellite down to 2° above the horizon).
The results provided here focus primarily on the fault detection and exclusion requirements for non-precision approach (NPA) and including the use of baro-aiding (shaded cells in Table 1b). To evaluate Each case, representing a collection of visible satellites observed at a space-time point, had been evaluated in a series of checks. First, the pre-fault detection may be tested using all available satellites. If the HIL meets the alert limit requirement, fault exclusion is tested. If each and every satellite can be uniquely identified should it fail, the post-fault detection of each of the satellite subsets that must operate without the failed satellite is assessed. To pass this test, the HIL of all these subsets must meet the alert limit requirements. If a particular satellite, should it fail, cannot be uniquely identified to a confidence level of better than 99.9%, we would attempt to exclude both that satellite together with the second satellite instigating the uncertainty in identification. If we can successfully exclude these two satellites, the subset of remaining satellites would be evaluated for post-fault detection to determine if it can meet the required alert limit. If these tests are not satisfied, the particular satellite geometry is deemed “unavailable”. Each of these “unavailable” cases can be placed in one of four different categories: A. Cannot pre-fault detect: The HIL exceeds the NPA alert limit of 555.6 meters with all the satellites; B. Cannot post-fault detect following single satellite identification: The attempt to isolate the worst-case satellite succeeded but the HIL of the remaining satellites exceeds the NPA alert limit in the post-fault detection test; C. Cannot post-fault detect following dual-satellite exclusion: The attempt to isolate the worst-case satellite failed but the attempt to exclude the worst-case satellite and the satellite with the characteristic vector closest to the former succeeded; however, after excluding the two satellites, the HIL of the remaining satellites exceeded the NPA alert limit in the post-fault detection test; D. Cannot exclude two satellites where needed: The attempt to exclude one or two satellites failed, implying the complication involving a third satellite; no attempt is made to exclude three satellites because the remaining subset of satellites is unlikely to be useful anymore. In each case, the H matrix contains the direction cosines of the visible satellites and the last row of the H matrix, [0 0 0.24 0], corresponds to the baro measurement. The third component associated with the vertical is the reciprocal of the ratio of the baro-altitude error to the GPS pseudorange error. Assume the former to be 50 meters and the latter to be 12 meters. In running the conventional RAIM-FDE tool in an exemplary text, for fault detection and exclusion with baro-aiding, 4127 “unavailable” cases were identified. These cases were broken down into the different types of categories described above according to the number of occurrences for each (see shaded column in Table 1b). The results in Table 2 show that the number of “unavailable” geometries can be reduced significantly with As for the other two categories, C and D, recall that Taking a closer look at the cases from Category B, by comparing standard RAIM-FDE with At the same time, To determine if a given geometry is “available”, the Lookup Table HIL was compared to the NPA Alert Limit of 555.6 m. As reasoned in the next section titled
A simple evaluation of the Lookup Table HIL approach was conducted by comparing it to one derived from a Monte Carlo approach. For each “unavailable” satellite geometry scenario, a set of weights was computed to obtain the lowest HIL based on the Lookup Table approach. Then, for that particular set of weights, a Monte Carlo run was made with 100,000 random samples and tested with a discrete-step incrementing fault to determine the HIL that yields a 10 The lookup table was constructed as a four-dimensional array with the four parameters used in the lookup scheme being: a) Number of satellites n from 5 to 9; b) Log √{square root over (c c) Correlation coefficient d) Slope of the satellite fault bias, s from 1.0 to 12.0 in steps of 1.0; The resulting table array contains 5×15×11×12 elements. Note that the chosen range for the slope component restricts its usefulness of generating HIL values mostly for NPA, and that range may have to be extended for other phases of flight. For a given number of satellites n, the remaining three parameters [log To evaluate the consistency between the Lookup Table HIL as compared to the Monte Carlo HIL, the difference between the two HIL values for each satellite geometry scenario were formed. Then, this difference was plotted against the Lookup Table HIL value. This plot shows a narrow scatter in the difference between the two HIL's, but there is a correlation between this difference and the absolute size of the HIL's. In the plot of For the HIL cases that are larger than this, the lookup table will extrapolate beyond the available cardinal points instead of interpolate between them. As a result, the Lookup Table. HIL estimate will likely be in error by a larger amount. It should be obvious that the discrepancy between the Lookup Table HIL and Monte Carlo HIL becomes exacerbated beyond about 1100 m. To be conservative, it may be preferable for all relevant data points in The Lookup Table used for the evaluations presented here appears to be more than adequate in handling availability evaluations associated with Non-Precision Approach where the alert limit is 555.6 m. For operational evaluations associated with higher alert limits, the Lookup Table needs to be extended, particular in the dimensions of c A novel modification to the standard RAIM-FDE algorithm called The general philosophy behind While the detailed drawings, specific examples and particular formulations given describe preferred and exemplary embodiments, they serve the purpose of illustration only. The inventions disclosed are not limited to the specific forms shown. For example, the methods may be performed in any of a variety of sequence of steps. The hardware and software configurations shown and described may differ depending on the chosen performance characteristics and physical characteristics of the computing devices. For example, the type of computing device, communications bus, or processor used may differ. The systems and methods depicted and described are not limited to the precise details and conditions disclosed. Furthermore, other substitutions, modifications, changes, and omissions may be made in the design, operating conditions, and arrangement of the exemplary embodiments without departing from the scope of the invention as expressed in the appended claims. In the two-dimensional Parity-Error space, in the presence of a fault, the sample statistical distribution moves up a slope line. The further its rise, as determined by the size of the fault, the more the distribution resembles a bivariate Gaussian. This approximation is useful for simplifying the Lookup Table HIL approach described earlier. Here, we will derive the covariance matrix of this bivariate Gaussian distribution:
For any two-dimensional distribution offset by a bias, we can form the following approximation. e=magnitude of total error Δx, Δy=components of random error X Since R
Now use this approximation to derive the terms of C
Do the same for the c However, since the parity space was transformed to normalize the distribution,
Finally, the cross term C Cov (p,e)
and
Citas de patentes
Otras citas Citada por
Clasificaciones
Eventos legales
Girar |